A carregar...
Publicação
Machine learning methods applied to the dots and boxes board game
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 5.51 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Pontos e Quadrados (Dots and Boxes na versão anglo-saxónica) é um jogo clássico
de tabuleiro no qual os jogadores unem quatro pontos próximos numa grelha para
criar o maior número possível de quadrados. Este trabalho irá inverstigar técnicas de
aprendizagem profunda e aprendizagem por reforço, que torna possível um programa
de computador aprender como jogar o jogo, sem nenhuma interação humana, e aplicar
o mesmo ao jogo Dots and Boxes; a abordagem usada no DeepMind AlphaZero será
analisada. O AlphaZero combina uma rede neural convolucional e o algoritmo Monte
Carlo Tree Search para alcançar um desempenho super humano, sem conhecimento
prévio, em jogos como o Xadrez, Go, e Shogi.
Os resultados obtidos permitem aferir sobre a adequação da abordagem ao jogo
Pontos e Quadrados.
Dots and Boxes is a classical board game in which players connect four nearest dots in a grid to create the maximum possible number of boxes. This work will investigate deep learning techniques with reinforcement learning to make possible a computer program to learn how to play the game, without human interaction, and apply it to the Dots and Boxes board game; the approach beyond DeepMind AlphaZero being taken as the approach to follow. AlphaZero makes a connection between a Convolutional Neural Network and the Monte Carlo Tree Search algorithm to achieve superhuman performance, starting from no a priori knowledge in games such as Chess, Go, and Shogi. The results obtained allow to measure the approach adequacy to the game Dots and Boxes.
Dots and Boxes is a classical board game in which players connect four nearest dots in a grid to create the maximum possible number of boxes. This work will investigate deep learning techniques with reinforcement learning to make possible a computer program to learn how to play the game, without human interaction, and apply it to the Dots and Boxes board game; the approach beyond DeepMind AlphaZero being taken as the approach to follow. AlphaZero makes a connection between a Convolutional Neural Network and the Monte Carlo Tree Search algorithm to achieve superhuman performance, starting from no a priori knowledge in games such as Chess, Go, and Shogi. The results obtained allow to measure the approach adequacy to the game Dots and Boxes.
Descrição
Palavras-chave
Adversarial search Machine learning Deep learning Reinforce-ment learning Dots and boxes Rede neural artificial Rede neural convolucional Jogos Alphazero Deepmind Jogos de tabuleiro Auto aprendizado